Here's a nice nomograph which will answer the question about the DMM's accuracy in % :
https://static.elektroda.pl/attach/AN82_6405927.png
Typically you're going to need much greater accuracy than the meter (general rule suggests 4x greater) in a standard in order to calibrate a meter. So if the meter is 4.5 digits with a good accurate A/D, that's between 0.01% and 0.005% precision and you'll need resistors of around 0.001% accuracy to be sure they're good enough to use for meter calibration. 0.25% resistors aren't even close, unless you hand-select them using a meter considerably more accurate than the one you're using for measurements (say, a 3458A).
I bought the PRC set (0.01%-0.02%) for about $70 and they're really not good enough for my 4.5 digit DMM, but then I only am using them for a spot check.
In general, the absolute value of the resistors is not that important, as long as they are reasonably close to nominal (0.1% is fine), and you know to a high degree of accuracy what the actual value is-- (thus, they need to be measured by a Cal-Lab, or someone with an in-cal high-spec meter or bridge). The TCR should be zero or as close to zero as you can manage, and the technology should be something that has excellent time drift specs (precision WW, hermetic metal foil, tantalum nitride).
There is nothing wrong with building a statistical array of tantalum nitride (Ta
2N) resistors-- selected for a zero TCR, and burned in at 125C for 90 days-- this will artificially age the resistors and the result will be exceptionally stable. Further trim of TCR can be done with a platinum thermistor (with appropriate shunt resistor), and further trim of the absolute value can also be done with an appropriate trim resistor in series or parallel with one of the resistors in the array. When you get all done, have a friend with a calibrated 8.5-digit meter measure them.